Radiation therapy systems and related methods

Abstract

A method is disclosed for establishing a biomechanical model (300) of an anatomical region (202) of a radiation therapy patient (200) for use in a radiation therapy system (100), comprising the steps of establishing a point-based or mesh-based model of the modeled anatomical region, associating static and / or dynamic features with points or mesh nodes (302) of the point-based or mesh-based model, the features including world-space coordinates of the points or mesh nodes, state variables representing spatiotemporal displacements of the anatomical region, and for each point or mesh node, at least one feature characterizing body tissue within the anatomical region, representing the model as a graph in a graph neural network system (500), and training the model for periodic or semi-periodic motions of the anatomical region.

Description

[Technical field]

[0001] The present disclosure relates to a radiation therapy system including a control system for controlling an ionizing radiation beam supervision system. In particular, the present disclosure relates to predicting motion, particularly periodic or semi-periodic motion, of an anatomical region to improve accuracy of dose delivery. [Background technology]

[0002] Treatment of disease, e.g., cancer treatment, using external beam radiation therapy involves exposing a patient to ionizing radiation such that the radiation energy is deposited within malignant cells in the patient's body. If a sufficient amount of energy is deposited, it can result in the destruction of DNA and subsequent death of the irradiated cells.

[0003] Ionizing radiation may be directly ionizing or indirectly ionizing. Direct ionizing radiation utilizes charged particles, such as electrons, protons, alpha particles, and heavy ions. Indirect ionizing radiation utilizes neutral particles, such as photons (X-rays and gamma rays) and neutrons. The present disclosure is applicable to both types of external beam radiation therapy, i.e., external beam radiation therapy using either directly ionizing radiation or indirectly ionizing radiation. In particular, the present disclosure is applicable to X-ray or proton radiation therapy.

[0004] Protons and other charged particles display a depth dose curve suitable for radiation therapy. Due to their relatively large mass, protons and other charged particles have little side scattering in tissue, and as a result, particle beams can be focused on malignant tissue and minimize dose side effects to surrounding healthy tissue. Particle beams can also target malignant tissue more precisely using the Bragg peak effect, i.e. the tendency of charged particles to deposit most of their energy in the very last region of the particle beam's trajectory.

[0005] US Patent Publication No. 2007 / 0133933 discloses a method for creating a proton therapy plan, the method including the steps of dividing a volume of interest into sub-volumes, applying dose constraints to the sub-volumes based, inter alia, on patient motion, finding one or more feasible configurations of a proton therapy system, and selecting a proton beam configuration that improves or optimizes one or more aspects of the proton therapy.

[0006] However, a problem associated with the proton therapy system of the '691 patent, as well as other prior art external beam radiation therapy systems, is that patient movement during treatment can make it difficult for the system to accurately deliver a radiation dose to an intended location within the irradiation target. Indeed, even if the patient is restrained, movement of the patient's internal organs, such as movement due to breathing and involuntary muscle activity, e.g., heartbeat, can still make it difficult for prior art radiation therapy systems to accurately deliver a radiation dose.

[0007] Thus, when creating a treatment plan using a prior art system, the uncertainty associated with patient and / or organ motion during treatment must be taken into account when planning dose delivery. In practice, this uncertainty can result in the system operator setting the system to deliver a lower overall dose to avoid damaging healthy tissue. This results in a less efficient treatment than would be the case if the uncertainty were not present.

[0008] In particular, in current radiation therapy practice, a relatively large margin is added to tumor volumes in the thorax and upper abdominal cavity to compensate for the effects of respiratory motion on tumor dose delivery, which can compromise the prescribed tumor dose as well as the treatment plan, thereby adversely affecting treatment outcomes and increasing the incidence of radiation-induced morbidity.

[0009] US Patent Publication No. 2009 / 0133633 discloses a motion compensation system for tracking and compensating for patient motion during medical image scanning. The system comprises an optical marker including an optically visible pattern, and first and second optical detectors arranged to digitally image the optically visible pattern along first and second lines of sight, respectively. The system further comprises a tracking engine configured to determine an object's pose in six degrees of freedom by analyzing images from the first and second optical detectors, and a controller interface configured to generate tracking information based on the pose and electronically transmit the tracking information to a scanner controller to enable compensation for object motion within the medical image scanner. US Patent Publication No. 2009 / 0133633 suggests that the disclosed method for tracking patient motion may be applied in a therapeutic context to track patient motion, such as due to respiration, to direct a therapeutic radiation beam to a diseased tissue region while avoiding surrounding healthy tissue.

[0010] However, while the system disclosed in US Patent No. 5,999,933 may be able to track patient movement to some extent by tracking the optical markers and assigning the movement of the optical markers to patient movement, the system does not take into account dynamic deformation or warping of the tissue area to which the optical markers are attached.

[0011] Biomechanical models for predicting patient motion are known in the art, for example, such as those discussed in non-patent literature 1 and 2. However, known biomechanical models, such as those discussed in the above papers, are computationally time-consuming and not suitable for real-time use during radiation therapy dose delivery. [Prior art documents] [Patent documents]

[0012] [Patent Document 1] US Patent Application Publication No. 2016 / 0144201 [Patent Document 2] International Publication No. 2014 / 116868

Non-licensed literature

[0013] [Non-licensed document 1] Patient-Specific Biomechanical Model for the Prediction of Lung Motion From 4-D CT Images (2014) Fuerst et al., IEEE transactions on medical imaging 34. 10.1109 / TMI.2014.2363611

Non-licensed Document 2

Non-licensed Document 4

Non-licensed Document 5

[0014] With the above problems in mind, it is an object of the present disclosure to provide a radiation therapy system and related systems and methods that can at least mitigate the above-mentioned problems and / or reduce uncertainties in target positioning during dose delivery.

[0015] Another object of the present disclosure is to provide a radiation therapy method and system that includes a beam monitoring system that allows for accurate dose delivery to the irradiation target.

[0016] Another object of the present disclosure is to provide a radiation therapy system with a beam monitoring system that can accurately track the irradiation target.

[0017] It is yet another object of the present disclosure to provide a radiation therapy system that can avoid irradiation of organs at risk.

[0018] A further object of the present disclosure is to provide a radiation therapy system that allows the delay time (latency) of the beam supervision system to be at least partially compensated for.

[0019] It is yet another object of the present disclosure to provide a method for establishing a model of an anatomical region of a radiation treatment patient for use in a radiation treatment system.

[0020] It is also an object of the present disclosure to provide a computer-implemented method for planning at least a portion of a radiation therapy treatment to be performed on an anatomical region of a radiation therapy patient. [Means for solving the problem]

[0021] The claimed invention is set out in the independent claims of the present application. Advantageous refinements and variations of the claimed invention are set out in the independent claims.

[0022] As mentioned above, the motion of an anatomical region may be modeled using a biomechanical model. The partial differential equations underlying such models may be solved in space and time using numerical computational methods, such as finite element analysis, finite differences or finite volume methods. The calculations may be carried out iteratively using time-step methods that allow the evolution of the model state to be obtained. However, as mentioned above, such calculations are slow and require significant computational resources, and therefore are not suitable for use in real time during radiation therapy dose delivery.

[0023] In the paper "Graph Network Based Simulators (GNS)", Graph Network Based Simulators (GNS) are discussed and identified as being able to accurately simulate a wide range of physical systems in which fluids, rigid solids, and deformable materials interact with each other.

[0024] In the above paper, a particle-based representation of a physical system is adopted, where each particle represents a state of the system. The physical dynamics are approximated by interactions between the particles, e.g., by exchanging energy and momentum between neighboring particles. Particle-based simulation can be viewed as message passing on a graph, where nodes correspond to particles and edges correspond to pairwise relations between particles, for which interactions are computed. A learnable simulator is a parametrized function approximator d θThe dynamics information is calculated using the function approximator, whose parameters can be optimized for some training purpose. The parameterized function approximator has three instruments: an encoder, a processor, and a decoder.

[0025] The encoder embeds the particle-based state representation as an input graph, where node embeddings are learned functions of the particle's states, directed edges are added to create paths between particle nodes that have some potential interactions, and graph-level embeddings can be added to represent global properties.

[0026] The processor computes interactions between the nodes through a learned message-passing step to generate an output graph.

[0027] The decoder extracts dynamics information from the nodes of the output graph.

[0028] In the paper "Non-Patent Document 4," a framework for learning mesh-based simulations using a graph neural network that implements an encoder-processor-decoder structure followed by an integrator is disclosed.

[0029] In the discussed framework, the state of the system at time t is described using a simulation mesh with nodes connected by mesh edges. Each node is associated with reference mesh-space coordinates that span the simulation mesh and additional dynamic quantities to be modeled. In systems where the mesh represents a moving and deforming surface or volume, the mesh contains, in addition to the fixed mesh-space coordinates, additional world-space coordinates that describe the dynamic state of the mesh in 3D space.

[0030] Given the current mesh, i.e., the mesh at time t, the framework learns a forward model of the dynamics of the mesh at time t+1.

[0031] The encoder encodes the current mesh into a multigraph. Mesh nodes become graph nodes and mesh edges become bidirectional mesh edges in the graph. World edges are added to the graph to allow learning external dynamics. World edges are edges created between spatially close nodes, excluding node pairs already connected by mesh edges. World edges allow modeling connectivity not only between adjacent mesh nodes, but also between nodes that are spatially close in world space but separated in mesh space. Features are encoded into the graph nodes and edges.

[0032] The relative displacement vector is encoded in mesh space and its norm is encoded on the mesh edges. The relative world-space displacement vector and its norm are encoded on both mesh and world edges. The remaining dynamic features as well as a one-hot vector indicating the node type are provided as node features.

[0033] The processor consists of multiple identical message-passing blocks, each containing a separate set of network parameters and applied sequentially to the output of the previous block, updating mesh edges, world edges and node embeddings.

[0034] After the final processing step of the processor, the latent node features are converted into one or more output features in the decoder. The output features are interpreted as (higher order) derivatives of the dynamic features and integrated to compute the dynamic quantities of the next stage. Additional output features may be used to make direct predictions of auxiliary quantities.

[0035] Finally, the output mesh nodes are updated to generate the mesh for time t+1.

[0036] The present disclosure is based on the discovery that a graph network system, if adopted and properly implemented, may be used as a basis for modeling and predicting periodic or semi-periodic motions and movements, in particular respiratory motions and movements, of an anatomical region containing an irradiation target in real time during radiation therapy dose delivery.

[0037] The predicted periodic or semi-periodic motion or movement of the anatomical region can be used to control the ionizing radiation beam during a radiation therapy session such that a desired dose delivery in the anatomical region is achieved. For example, the alignment of the beam can be controlled based on the predicted motion or movement such that the irradiation target is continuously tracked. Additionally or alternatively, the beam can be blocked or stopped when the predicted motion or movement of the anatomical region indicates that the irradiation target is likely to move outside the range or trajectory of the ionizing radiation and / or when the predicted motion or movement indicates that the ionizing radiation beam is likely to intersect with an organ at risk. Thus, the present disclosure is applicable to radiation therapy systems that can control the alignment of the ionizing radiation beam as well as radiation therapy systems that cannot control the alignment of the ionizing radiation beam, in the former case, beam tracking functionality and / or beam blocking or beam stopping functionality can be used, and in the latter case, only beam blocking or beam stopping functionality can be used.

[0038] According to a first aspect, the present disclosure provides a method of establishing a model of an anatomical region of a radiation therapy patient for use in a radiation therapy system, the method comprising: (a) establishing a point-based or mesh-based representation or model of the anatomical region; (b) associating static and / or dynamic features with points or mesh nodes of said point-based or mesh-based representation or model, said features including world-space coordinates of said points or mesh nodes, state variables representing space-time displacements of said anatomical region, and for each point or mesh node, at least one feature characterizing body tissue within said anatomical region; (c) representing the point-based or mesh-based representation or model as a graph in a graph neural network system; (d) training the point-based or mesh-based representation or model for periodic or semi-periodic motion of the anatomical region; Includes.

[0039] The step of establishing a point-based or mesh-based representation or model of the anatomical region typically involves modeling the anatomical region as interconnected points or as a mesh comprising interconnected mesh nodes, and the step of representing the point-based or mesh-based representation or model as a graph in a graph neural network system typically involves employing the particle-based or mesh-based representation of the anatomical region such that each particle or mesh node represents a state of the system and physical dynamics are approximated by interactions between the particle or mesh nodes.

[0040] The step of training the point-based or mesh-based representation or model on periodic or semi-periodic motion of the anatomical region typically comprises training the graph neural network system on periodic or semi-periodic motion of the anatomical structure, i.e. training the graph neural network system to predict periodic or semi-periodic motion of the anatomical region.

[0041] The at least one feature characterizing the body tissue within the anatomical region may advantageously include at least one variable characterizing an interaction between the anatomical region and the ionizing radiation beam. In particular, the at least one feature characterizing an interaction between the anatomical region and the ionizing radiation beam may include at least one of linear stopping power, relative stopping power, radio density, and radiopacity.

[0042] Representing the graph as a graph in a graph neural network system may include establishing graph nodes, graph edges, and graph global elements having characteristics reflecting static and dynamic variables of the biomechanical model.

[0043] The graph network system may comprise an input, a graph processing neural network, an updater, and an output, and the graph processing neural network may comprise an encoder, a processor, and a decoder.

[0044] The encoder may be configured to receive a current state of the mesh-based model from the input and to encode an input graph from the current state of the model, and the processor may be configured to process the input graph and provide an output graph having updated node and edge embeddings.

[0045] Further, the decoder may be configured to decode the output graph to extract output features corresponding to the static and dynamic variables of the biomechanical model, and the updater may be configured to update the output features to generate an estimated future state of the mesh-based biomechanical model and provide the estimated state to the output.

[0046] Training the model for periodic or semi-periodic motion of the anatomical region may include applying a loss function to the estimated states provided by the updater and to a real data set of an anatomical region corresponding to the modeled anatomical region.

[0047] The real-world data set may include at least one of a four-dimensional computed tomography (4D-CT) scan, a four-dimensional magnetic resonance imaging (4D-MRI), and ultrasound imaging.

[0048] The real data set may include a progression of images covering the modeled anatomical region during a respiratory cycle.

[0049] The point-based or mesh-based model may incorporate at least one of a linear elastic model and a hyperelastic model adapted to the physical properties of the anatomical region.

[0050] The method comprises: dividing the anatomical region into a first anatomical sub-region displaying a first set of physical properties and a second anatomical sub-region displaying a second set of physical properties different from the first set of physical properties; assigning, for the first anatomical sub-region, to the point-based or mesh-based model a first point-based or mesh-based sub-model adapted to the first set of physical properties; and assigning to the point-based or mesh-based model, for the second anatomical sub-region, a second point-based or mesh-based sub-model that is adapted to the second set of physical properties.

[0051] The first and second anatomical sub-regions may share a common boundary, and the method may include assigning common boundary conditions at the common boundary to the first and second point-based or mesh-based sub-models.

[0052] Such a graph network system may be implemented according to any of the papers in Non-Patent Document 3 and Non-Patent Document 4 mentioned above.

[0053] According to a second aspect, the present disclosure provides a computer-implemented method for planning at least a portion of a radiation therapy treatment to be performed on an anatomical region of a radiation therapy patient, the method comprising: Establishing a biomechanical model of said anatomical region (202) according to the method of the first aspect; and assigning to each point or mesh node at least one of a planned dose and a maximum allowed dose.

[0054] According to a third aspect, the present disclosure provides a method of providing a radiation therapy system comprising at least one beam source arranged to deliver a beam of ionizing radiation onto a radiation target located within an anatomical region of a radiation therapy patient, the method comprising: establishing a biomechanical model of the anatomical region; training said model for periodic or semi-periodic motion of said anatomical region; implementing the trained model in a graph network system implemented in a control system of the radiation therapy system; establishing a current state of the trained model in the graph network system based on data relating to a current position in space and / or a current orientation of the anatomical region; establishing an estimated future state of the trained model in the graph network system based on the current state of the trained model; and controlling the ionizing radiation beam of the at least one beam source based on the estimated future state.

[0055] The step of controlling the ionizing radiation beam of the at least one beam source may include controlling an alignment of the at least one beam source such that the ionizing radiation beam tracks the irradiation target based on the estimated future state.

[0056] Additionally or alternatively, the step of controlling the ionizing radiation beam of the at least one beam source may include blocking the ionizing radiation beam or stopping the at least one beam source when the estimated future state indicates that the irradiation target is about to move outside the trajectory of the ionizing radiation beam.

[0057] Additionally or alternatively, the step of controlling the ionizing radiation beam of the at least one beam source may include blocking the ionizing radiation beam or stopping the at least one beam source when the estimated future state indicates that there is a risk that the trajectory of the ionizing radiation beam will intersect with an organ at risk.

[0058] Blocking the radiation beam may be achieved by applying a gating function to a beam supervision system, ie, a subsystem of the radiation therapy system that controls the at least one beam source.

[0059] Said periodic or semi-periodic motion may typically be a respiratory motion.

[0060] The step of establishing the biomechanical model may include establishing a point-based or mesh-based model of the anatomical region and associating static and / or dynamic features with points or mesh nodes of the point-based or mesh-based model. The features may include world-space coordinates of the points or mesh nodes and state variables representing the space-time displacement of the anatomical region. The state variables may be a displacement field, i.e. an assignment of displacement vectors to all points or mesh nodes in the anatomical region that have been displaced from one state to another.

[0061] Implementing the trained model in the graph network system may include representing the trained model as a graph in the graph network system by establishing graph nodes, graph edges, and graph global elements having features reflecting static and dynamic variables of the trained model.

[0062] The step of establishing a current state of the trained model may include relating the current state to data regarding the current position and / or orientation in space of the anatomical region provided by a monitoring system of the radiation therapy system configured to monitor the anatomical region.

[0063] The monitoring system may be configured to monitor at least one of a position and / or orientation in space of a boundary region of the anatomical region and a position and / or orientation in space of a fiducial marker in the anatomical region.

[0064] The variables may include at least one variable characterizing an interaction between the anatomical region and a beam of ionizing radiation, such as linear stopping power and / or radiopacity. The variables may additionally or alternatively include at least one variable characterizing at least one of a planned radiation dose, a maximum tolerable radiation dose, and an accumulated radiation dose.

[0065] Providing the estimated future state of the trained model may include updating the variables characterizing the accumulated radiation dose, and controlling the position and / or alignment of the at least one beam source and / or blocking the ionizing radiation beam may include instructing a beam supervision system of the radiation therapy system to position and align the at least one beam source such that the irradiation target receives the planned radiation dose without exceeding the maximum tolerable radiation dose for other parts of the anatomical region.

[0066] According to a fourth aspect, the present disclosure provides a radiation therapy system, the radiation therapy system comprising: at least one beam source positioned to provide a beam of ionizing radiation to an irradiation target located within the anatomical region; a beam supervision system configured to control the at least one beam source; and a monitoring system arranged to monitor said anatomical region and, optionally, said beam supervision system; a control system comprising a computer implemented graph network system in which a model of the anatomical region, trained on periodic or semi-periodic motion of the anatomical region, is implemented; the graph network system is configured to establish a current state of the trained model based on data received from the monitoring system relating to a current position and / or a current orientation in space of the anatomical region, and further configured to establish an estimated future state of the trained model based on the current state; The control system is configured to instruct the beam supervision system to control the ionizing radiation beam of the at least one beam source based on the estimated future state.

[0067] The control system may be configured to instruct the beam supervision system to control the ionizing radiation beam of the at least one beam source based on the estimated future state and to cause the at least one beam source to track the irradiation target.

[0068] Alternatively or additionally, the control system may be configured to control the ionizing radiation beam of the at least one beam source based on the estimated future state, and to instruct the beam supervision system to block the ionizing radiation beam of the at least one beam source or to stop the at least one beam source if the estimated future state indicates that the irradiation target is about to move outside the trajectory of the ionizing radiation beam.

[0069] Alternatively or additionally, the control system may be configured to control the ionizing radiation beam of the at least one beam source based on the estimated future state, and to instruct the beam supervision system to block the ionizing radiation beam of the at least one beam source or to stop the at least one beam source when the estimated future state indicates that there is a risk that the trajectory of the ionizing radiation beam will intersect with an organ at risk.

[0070] The trained model may be a point-based or mesh-based model, comprising static and / or dynamic features associated with points or mesh nodes of the point-based or mesh-based model, the features including world-space coordinates of the points or mesh nodes and state variables representing spatio-temporal displacements of the anatomical region.

[0071] The graph network system may be configured to represent the trained model as a graph including graph nodes, graph edges, and graph global elements having features that reflect the static and dynamic variables of the trained model.

[0072] The graph network system may be configured to associate the established current state of the trained model with data received from the monitoring system.

[0073] The monitoring system may be configured to monitor at least one of a position and / or orientation in space of a boundary region of the anatomical region and a position and / or orientation in space of a fiducial marker in the anatomical region.

[0074] The variables may include at least one variable characterizing an interaction between the anatomical region and a beam of ionizing radiation, and at least one variable characterizing at least one of a planned radiation dose, a maximum tolerable radiation dose, and an accumulated radiation dose. For each estimated future state of the trained model, the graph network system may be configured to update the variable characterizing the accumulated radiation dose, and the control system may be configured to instruct the beam supervision system to control a position and / or alignment of the at least one beam source such that the irradiation target receives a planned radiation dose without exceeding the maximum tolerable radiation dose for other parts of the anatomical region.

[0075] According to a fifth aspect, the present disclosure provides a radiation therapy system, the radiation therapy system comprising: at least one beam source positioned to deliver a beam of ionizing radiation to an anatomical region containing an irradiation target; a beam supervision system configured to control the position and / or alignment of the at least one beam source; and a monitoring system arranged to provide updated current data regarding the position and / or orientation in space of said anatomical region; a control system including a computer implemented graph network system in which a trained model of the anatomical region for periodic or semi-periodic motion of the anatomical region is represented as a graph including nodes, edges and global elements, and including static attributes reflecting static characteristics of the anatomical region and dynamic attributes reflecting dynamic properties of the anatomical region, the control system comprising: providing a current representation or status of the anatomical region based on information received from the monitoring system; processing the current representation or state in the graph network system to provide an estimated future representation or state of at least one of the anatomical regions; configured to control the ionizing radiation beam of the at least one beam source based on the at least one estimated future representation or state of the anatomical region.

[0076] Thus, the radiation therapy system provides the control system with predicted future values ​​of the dynamic characteristics of the anatomical region, enabling the beam supervision system to adapt the alignment of the beam to take into account predicted movements and deformations of the anatomical region, in particular movements and deformations of the anatomical region due to respiratory activity of the radiation therapy patient, thereby enabling the radiation therapy system to accurately track the irradiation target while taking into full account surrounding tissues, in particular organs at risk. This also allows delay times of the beam supervision system, in particular delay times of the beam guidance subsystem of the beam supervision system, to be at least partially compensated.

[0077] Additionally or alternatively, if the predicted future values ​​of the dynamic characteristics of the anatomical region indicate that the ionizing radiation beam may intersect with an organ at risk, the control system may be configured to block the ionizing radiation beam or to stop the at least one beam source.

[0078] Additionally or alternatively, if the predicted future values ​​of the dynamic characteristics of the anatomical region indicate that the irradiation target is likely to move outside the trajectory of the ionizing radiation beam, e.g. due to movement of the anatomical region that would move the irradiation target outside the range of the at least one beam source, the control system may be configured to block the ionizing radiation beam or stop the at least one beam source.

[0079] The graph network system may preferably be configured to provide an estimated future representation or state of at least one of the anatomical regions repeatedly during dose administration, preferably with a periodicity in the range of 20-100 ms, e.g. 50 ms.

[0080] The graph network system may comprise an encoder configured to encode an input graph from the current representation of the anatomical region; a processor comprising one or more graph network blocks and trained to process the input graph to perform message passing along graph edges and provide an output graph having updated node and edge embeddings; and a decoder configured to decode the output graph to extract at least one output feature from the output graph.

[0081] Such a graph network system may be implemented according to any of the papers in Non-Patent Document 3 and Non-Patent Document 4 mentioned above.

[0082] The features may advantageously include features characterizing body tissue within the anatomical region. The tissue features may include line or relative stopping power and / or radio density or radiopacity.

[0083] The features may also include at least one of a planned dose, an accumulated dose, and a maximum allowed dose associated with each point or mesh node.

[0084] Information regarding the position and / or alignment of the beam source may be provided to the control system. In addition to being based on information received from an imaging system, the current representation of the anatomical region may also be based on information regarding the alignment of the beam source. Thus, if the dynamic features include absorbed dose, the estimated future representation of the anatomical region is also updated with respect to the spatial distribution of absorbed dose, allowing dose delivery and beam alignment to be controlled based on the difference between planned dose and absorbed dose.

[0085] The representation of the anatomical region may be provided as a mesh including mesh nodes and mesh edges, where the static and dynamic attributes may reflect static and dynamic characteristics of the mesh nodes.

[0086] According to a sixth aspect, the present disclosure provides a computer implemented control system for controlling an ionizing radiation beam supervision system to track an irradiation target within an anatomical region of a radiation therapy patient during dose delivery and / or to block the ionizing radiation beam when the ionizing radiation beam is at risk of intersecting an organ at risk or when the irradiation target is about to move outside the trajectory of the ionizing radiation beam, the control system comprising a computer implemented neural network system for processing data representing the irradiation target, wherein the anatomical region (including the irradiation target) is bounded by a mesh edge E. M A simulation mesh M containing nodes V connected by t = (V, E M ) is modeled in the neural network system, and each node

number

number

number

[0087] According to a seventh aspect, the present disclosure provides a computer-implemented method for modeling periodic or semi-periodic motion of an anatomical region including an irradiation target, the method comprising: A mesh edge E describing the anatomical region M A simulation mesh M containing nodes V connected by t = (V, E M ), wherein each node

number

number

number

number

number

number

number

number

number

number

number

number

number

number

number

number

number

number

number

number

[0088] According to an eighth aspect, the present disclosure provides a biomechanical model of an anatomical region including an irradiation target, trained for periodic or semi-periodic motion in a trained graph network system.

[0089] The features of any one of the above-described aspects may be combined with the features of any one of the other aspects.

[0090] The disclosed systems and methods include an anatomical biomechanical model that represents and simulates a patient's internal conditions, and a graph neural network that is trained on the simulated biomechanical data to learn the patient's internal conditions.

[0091] The biomechanical model can be used to simulate the movement and deformation of internal organs during breathing, which can be used to predict the internal conditions of a patient during radiation therapy or surgery.

[0092] The biomechanical model may rely on a collection of meshes that are three-dimensional (3D) in space, each representing a single anatomical structure, or a collection of point clouds that define various physical domains (mesh-free), such as organs and tissues.

[0093] The biomechanical model may be mathematically defined by a set of dynamic and constitutive equations, as well as boundary conditions, parameters and initial values.

[0094] The biomechanical model may be solved in space and time to generate a series of estimates of three-dimensional deformation fields representing the dynamic internal motion of anatomical structures, which may be used to predict the location of radiation deposition.

[0095] The biomechanical model may be solved in space and time using numerical methods such as finite element analysis, finite differences or finite volume methods.

[0096] The calculations may be carried out iteratively using a time-stepping method that makes it possible to obtain an almost real-time evolution of the model state during radiation delivery, where real-time means that said model state is updated frequently enough to allow changes in the internal anatomical structures to be captured and recorded with sufficient resolution.

[0097] High-resolution simulations are often very slow and, in accordance with the present disclosure, learned models may be used instead to provide faster predictions. Biomechanical modeling (mesh-based or particle-based) may be used to generate data for training neural networks, which may in turn be used in real-time calculations of the biomechanics of patient tissue.

[0098] The computational model may predict the three-dimensional spatial coordinates of target and risk tissues, as well as spatially varying tissue parameters, such as Hounsfield Units (HU) and / or Relative Stopping Power (RSP), at a required time step t in the future, ranging from milliseconds to seconds. Spatially varying quantities, such as HU and RSP, may be extracted from a baseline CT image of the patient prior to simulation and assigned to respective nodes in the three-dimensional anatomical model. Since the spatial resolution of the mesh or point cloud may be lower than the original CT image, each data node may represent a neighborhood of values.

[0099] The term "model" in this context refers to a model that can simulate (and predict) displacements over space and time, given a set of input parameters and a description of the system dynamics. The term "3D anatomical model" may refer to a geometric representation of an anatomical structure or the entire patient body. The anatomical region may be the chest, abdomen, head and neck or brain. The anatomical structure may be a solid cancerous or non-cancerous tumor, an organ, a tissue, or a combination thereof. The proposed motion estimation method can be applied to any structure that undergoes periodic or semi-periodic translation, rotation, and / or deformation.

[0100] [Biomechanical modeling] A dynamic model of a physical domain can be broadly specified by a system of partial differential equations F with source terms f, boundary conditions g and h, and model parameters p. The model may also include a set of constitutive equations and intermediate parameters. Model state variables

number

number

[0101] In the disclosed system and method, the state variable

number

[0102] The mathematical model described above is able to represent, within a certain accuracy range, the relevant dynamics of the physical domain (in this case,

number

[0103] Each distinct physical domain may be modeled by a different set of equations, and therefore the construction of such a model may require expertise in the physics and mechanics of the corresponding domain. The requirement is that the model is capable of generating spatially and temporally resolved displacements of the domain nodes. Identical boundary conditions may be applied to domains that share boundaries.

[0104] The biomechanical model may be the mathematical model above or the state variables from time t to t+1.

number

[0105] -Elastic model- There are many mathematical models that describe the behavior of materials, including the following: Linear Elastic Model: A linear elastic material describes the deformation of an object under an external or internal load as linearly proportional to the load. Hooke's law describes the behavior of the material and relates the unknown stress to the strain. Hyperelastic models: such materials not only exhibit nonlinear material behavior, but also large deformations and stress-strain relationships are derived from the strain energy density function. Examples of hyperelastic material models are the Neo-Hookean and Mooney-Rivlin models.

[0106] The biomechanical model may incorporate any of these mathematical models.

[0107] -Mesh-based and meshless methods- The biomechanical 3D anatomical model may be represented either as a tetrahedral mesh or as a cloud of points / particles, and different numerical methods may be used to solve the fundamental system of partial differential equations.

[0108] -Meshless / particle-based method- Meshless methods may utilize unstructured point clouds for spatial discretization. The collocation of nodes within a domain may be arbitrary or may follow certain rules, for example ensuring that a portion of the domain contains a higher node density, allowing for higher solution accuracy in that subdomain.

[0109] -Mesh-based method- Numerical simulations in biomechanics are usually solved by the Finite Element Method (FEM). Independent organ / tissue domains may be discretized into a finite set of interconnected nodes and cells, for which finite element analysis solutions may be computed. Meshes may be generated from organ and tumor partitions, dividing the structures into small connected tetrahedral elements.

[0110] -domain- The domain may be considered as a physical continuum that can be described by the same set of equations, parameters, and constraints. The physical domain of the biomechanical model may be the anatomical structure of the subject, which typically includes the tumor to be treated, the organ in which the tumor is formed, and the surrounding organs at risk (OARs). They may also include non-OARs and the external surface of the body.

[0111] The domain may be partitioned into a finite number of cells or nodes, resulting in either a mesh or a point cloud, which may serve as the basis for model discretization in methods for numerically solving the biomechanical model.

[0112] -Boundary conditions- Boundary conditions may be defined prior to simulation to constrain the biomechanical model to a unique solution. In case of mesh-based models, boundary conditions may be defined on each domain (boundary) facet. Boundary conditions may specify either the desired value of the state variables on the boundary (Dirichlet boundary conditions), or the gradient of the state variables (Neumann boundary conditions), or a combination of both (Robin, mixed, Cauchy boundary conditions). Since the state variables are displacements, which are vector quantities, the boundary conditions are also vector quantities. Thus, the boundary conditions can be found by calculating the deformation from one image (reference image) to the other image (movement image). Therefore, to obtain the boundary conditions from the images, at least two patient images from the same imaging study, taken at different times, are required.

[0113] Boundary conditions may also be obtained from alternative data sources other than imaging, for example from surveillance sensors, in which case the boundary conditions are typically used to restrict sub-regions of the boundary.

[0114] The boundary conditions may be patient-specific, ie, derived from the patient's own image data, or may be derived from other patient data.

[0115] -The power inside- Forces and pressures on internal organs, driven for example by muscle movement, respiration, boluses (IV boluses), bladder infusions, etc., may be embedded into the biomechanical model by including forcing terms (also called source terms) in the differential equations. These forces may be used to further constrain the model to behave in a physiological (as well as patient-specific) manner and may therefore be tailored to the application.

[0116] -Solving the biomechanical model- The biomechanical model may be solved using any suitable FEM software.

[0117] [Graph network framework] The disclosed method and system uses a graph network framework or system to predict the future state of a biomechanical model, for example using a framework based on the teachings of the paper by E.

[0118] Methods and systems according to the present disclosure also use a graph network framework or system to learn the dynamics of a biomechanical model.

[0119] Within a graph network framework or system, a graph typically consists of three elements:

number

number

[0120] The primary unit of computation in a graph network framework or system is the graph network block (GN block), which is a "graph-to-graph" module that takes in a graph as input, performs computations on the structure, and returns a graph as output.

[0121] A graph network block consists of three subfunctions: edges E, nodes V, and global attributes

number

[0122] The implementation of the subfunctions may be based on DeepMind's Sonnet library (https: / / github.com / deepmind / sonnet).

[0123] If the 3D biomechanical model is mesh-based, graph nodes and edges may be specified to represent the nodes and vertices of the mesh, respectively. The dynamics of the system, i.e., state variables and their derivatives, may be embedded in the edges of the graph. The processor of the graph network block takes the input graph, updates the dynamics defined on the edges, and generates the output graph.

[0124] The features received by each edge determine the physical interactions and properties from the biomechanical model that are encoded in the network.

[0125] The graph edges are the relative displacement fields

number

number

[0126] The differential operator is a subfunction (f M ) embedded in the edges of the graph. These functions take as input the edge features (e.g., derivatives and forces) and the sending and receiving node features (coordinates, material properties, etc.) and output the higher order derivatives of the nodes. These are then approximated by the dynamic variables

number

[0127] Graph nodes also store functions (f v ) may be embedded.

number

number

number

[0128] [Table 1]

[0129] The table below describes graphical features of the system and method according to one aspect of the present disclosure.

[0130] [Table 2]

[0131] For the first-order model, the mechanical quantity x at time t+1 i is obtained by integrating the derivative at time t according to the following formula:

number

[0132] Therefore, for the first-order model, the mechanical quantity x i is updated using the first derivatives output by the graph network.

[0133] For the quadratic model, the mechanical quantity x at time t+1 i is obtained according to the following formula:

number

[0134] For the quadratic model, x' i is embedded as a node feature, and x" i is the mechanical quantity x i is used to integrate [Brief description of the drawings]

[0135] [Figure 1] FIG. 1 is a schematic diagram illustrating an embodiment of a radiation treatment system according to the present disclosure. [Diagram 2] FIG. 1 is a schematic diagram illustrating an embodiment of a radiation treatment system according to the present disclosure. [Diagram 3] FIG. 1 illustrates anatomical regions of a radiation therapy patient. [Figure 4] FIG. 4 shows a mesh-based model of the anatomical region according to FIG. 3. [Diagram 5] FIG. 5 shows a graph encoded from the mesh-based model according to FIG. 4. [Figure 6] FIG. 1 illustrates a graph network system. [Figure 7] FIG. 2 illustrates a graph network block. [Figure 8] FIG. 1 illustrates an embodiment of a method for establishing a biomechanical model of an anatomical region of a radiation treatment patient for use in a radiation treatment system. [Figure 9] FIG. 1 illustrates an embodiment of a computer-implemented method for planning at least a portion of a radiation therapy treatment to be performed on an anatomical region of a radiation therapy patient. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0136] The present invention will now be described in more detail with reference to the accompanying drawings.

[0137] An embodiment of a radiation therapy system according to the present disclosure and how such a system is provided will be described in more detail below in the context of a radiation therapy system 100 configured to predict and compensate for respiratory motion of a radiation therapy patient during dose delivery, although it should be understood that the following teachings may be applied to predict and compensate for any type of periodic or semi-periodic motion.

[0138] Figures 1 and 2 are schematic diagrams illustrating an embodiment of a radiation therapy system 100 according to the present disclosure. The radiation therapy system 100 is configured to irradiate an irradiation target in an anatomical region of a radiation therapy patient. Figure 3 illustrates a radiation therapy patient 200 having an anatomical region 202, in this case the breast, in which there is an irradiation target 204. In Figures 1 and 2, the anatomical region and the irradiation target are indicated generally by the reference numerals 202 and 204, respectively.

[0139] The anatomical region 202 includes different anatomical structures with different tissue properties, some of which are organs at risk OARs, i.e. structures or organs that should preferably not be irradiated during radiation therapy. The irradiation target 204 may typically be a tumor.

[0140] Providing the radiation therapy system 100 involves establishing, prior to a radiation therapy session, a three-dimensional mesh-based biomechanical model of the anatomical region 202 including mesh nodes and mesh vertices, and associating each mesh node with static and dynamic variables within or at that mesh node that characterize properties of the anatomical region 202, such that the state of the anatomical region 202 can be represented by a collection of mesh nodes that encode properties within a local region of space of the anatomical region 202.

[0141] 4 illustrates a biomechanical model 300 of the anatomical region 202 disclosed in FIG. 3, including mesh nodes 302 and mesh vertices 304, where each mesh node 302 is associated with a number of static and dynamic variables. The static and dynamic variables are related to the world-space coordinates of the mesh node 302.

number

number

number

number

[0142] Advantageously, the static and dynamic variables may also include at least one variable characterizing the interaction between the anatomical region 202 and the ionizing radiation beam within or at each mesh node 302 or voxel, such as linear stopping power or radiopacity (e.g., expressed as Hounsfield units). The static and dynamic variables may also include variables characterizing the planned radiation dose and / or the maximum tolerated radiation dose within or at each mesh node 302. The static and dynamic variables may also include variables characterizing the accumulated radiation dose within or at each mesh node 302.

[0143] The biomechanical model 300 may in principle be based on any mathematical model of an anatomical region that allows simulating the movements and deformations of organs and tissues in the anatomical region 202. The biomechanical model may be mathematically defined by a set of dynamic and constitutive partial differential equations, as well as boundary conditions, parameters and initial values, which equations may be solved in space and time to generate a series of estimates of a three-dimensional deformation field that represents the dynamic internal motion of the anatomical region 202. Typically, such a solution involves solving the equations in space and time using numerical computational methods such as finite element analysis, finite differences or finite volume methods. Such computational methods may be carried out iteratively using time-step methods, which allow to obtain an evolution of the model state that is updated with a sufficiently high frequency, allowing the changes in the internal anatomical structures to be captured and recorded with sufficient resolution. Different sub-regions of the anatomical region 202 may be modeled using different mathematical models. In particular, different sub-regions may be modeled by different sets of equations, taking into account the different tissue types present in the biomechanical model 300.

[0144] For example, establishing the biomechanical model 300 may include dividing the anatomical region 202 into at least a first anatomical sub-region 202a displaying a first set of physical properties and a second anatomical sub-region 202b displaying a second set of physical properties different from the first set of physical properties (see FIG. 1 ). For the first anatomical sub-region 202a, the mesh-based model 300 may be assigned a first mesh-based sub-model adapted to the first set of physical properties, and for the second anatomical sub-region 202b, the mesh-based model 300 may be assigned a second mesh-based sub-model adapted to the second set of physical properties.

[0145] The first and second anatomical sub-regions 202a, 202b may share a common boundary, and the method of establishing the biomechanical model 300 may include assigning common boundary conditions at the common boundary to the first and second point-based or mesh-based sub-models.

[0146] There are numerous mathematical models that describe the behavior of materials. These models include the linear elastic model, which describes the deformation of an object under an external or internal load as linearly proportional to the load. In these models, Hooke's law describes the material behavior and relates the unknown stress and strain. Known mathematical models also include the hyperelastic model, where the material exhibits nonlinear material behavior as well as large deformations and stress-strain relationships that are derived from the strain energy density function. Examples of hyperelastic material models include the Neo-Hookian model and the Mooney-Rivlin model.

[0147] The biomechanical model 300 may incorporate any one of these mathematical models.

[0148] In general, the biomechanical model 300 may be defined by a system of partial differential equations F with source terms f, boundary conditions g and h, and model parameters p.

number

[0149] Model State Variables

number

[0150] Providing the radiation therapy system 100 further includes training a model for periodic or semi-periodic motion, the radiation therapy system 100 being configured to predict and compensate for respiratory motion in this embodiment.

[0151] Preferably, this training involves representing the model 300 as a graph in a graph network system and applying a loss function to the estimated future states provided by the graph network system.

[0152] Representing the model 300 as a graph in a graph network system involves establishing graph nodes, graph edges, and graph global elements that have characteristics that reflect the static and dynamic variables of the biomechanical model 300 .

[0153] Graph networks are known, for example, from the aforementioned papers, Non-Patent Document 3 and Non-Patent Document 4, and the graph network related to the radiation therapy system 100 may be advantageously implemented starting from the teachings of any of these papers.

[0154] 5 is a schematic diagram of a graph 400 representing the biomechanical model 300 shown in FIG. 4, the graph including graph nodes 402 and graph edges 404. In this embodiment, the graph nodes 402 and graph edges 404 may be identified as representing the mesh nodes 302 and mesh vertices 304, respectively, and may represent, for example, the displacement fields

number

[0155] FIG. 6 is a schematic diagram illustrating the main components of a graph network system or framework 500 implemented in this embodiment of the radiation therapy system 100.

[0156] The graph network system 500 comprises an input 502, a graph processing neural network 504, an updater 506, and an output 508. The graph processing neural network 504 comprises an encoder 510, a processor 512, and a decoder 514.

[0157] The encoder 510 derives the current state M of the mesh-based biomechanical model 300 from the input 502. t , and the current state M of the biomechanical model 300 is t from the input graph G 0 The method is configured to encode the

[0158] The processor 512 receives an input graph G 0 , and output graph G with updated node and edge embeddings. MThe processor 512 includes one or more graph network blocks 516 (see FIG. 7) that take a graph as input, perform computations on the graph structure, and return a graph as output. Each graph network block includes three sub-functions, one for each of the graph edges, graph nodes, and graph global elements. The sub-functions may be implemented by potentially using any class of neural network. Each of the sub-functions may be implemented, for example, using a multi-layer perceptron (MLP), a recurrent neural network (RNN), a long short-term memory (LSTM) recurrent neural network, or a convolutional neural network (CNN).

[0159] The implementation of the subfunction can be based on DeepMind's Sonnet library (https: / / github.com / deepmind / sonnet).

[0160] The decoder 514 generates an output graph G that extracts output features corresponding to the static and dynamic variables of the biomechanical model 300. M The method is configured to decrypt the

[0161] The updater 506 updates the output features to estimate the future state M of the mesh-based biomechanical model 300. t+1 Generate and estimate the state M t+1 to an output 508.

[0162] Training the biomechanical model 300 preferably involves applying a loss function to the estimated states provided by the updater and to a real data set of the anatomical region corresponding to the modeled anatomical region 202. The real data set may be based on a four-dimensional computed tomography (4D-CT) scan of the anatomical region corresponding to the modeled anatomical region 202. For this purpose, the Hounsfield units, which give the opacity of tissue material to X-rays, should preferably be included as a variable in the model 300. Alternatively or additionally, the training may be based on a data set obtained from four-dimensional magnetic resonance imaging (4D-MRI) or ultrasound imaging.

[0163] To train the respiratory motion, the data set may preferably be based on an evolution of images covering the modeled anatomical region 202 during the respiratory cycle. A minimum of two image data sets of the same patient, each capturing a different respiratory pattern, can be used to train the model. However, preferably images from a larger patient population showing different respiratory patterns (inter- and intra-patient variability) can be used.

[0164] 1 and 2, the radiation therapy system 100 comprises at least one beam source 102 arranged to provide an ionizing radiation beam 104. The radiation beam 104 may be, for example, a particle beam, e.g., a proton beam, or a photon beam, e.g., an X-ray beam.

[0165] The radiation therapy system 100 further comprises a beam supervision system 106 configured to control the position and alignment in space of each beam source 102. This control may preferably include controlling three variables defining the position of each particle beam source, e.g. represented by Cartesian coordinates x, y, z, and / or variables defining the pitch, yaw, and / or roll of the beam source, e.g. represented by rotation angles measured about orthogonal pitch, yaw, and / or roll axes, as shown diagrammatically in Figures 1 and 2.

[0166] The radiation therapy system 100 also comprises a control system 110 configured to control the beam supervision system 106 to track the irradiation target 204. To this end, the control system 110 comprises a computer-implemented graph network system or framework 500 in which the above-mentioned trained biomechanical model 300 is implemented. The graph network system 500 may advantageously be of the same kind as that used to train the model 300.

[0167] The radiation therapy system 100 also includes a monitoring system 120 arranged to monitor the anatomical region 202 and receive updated data regarding at least one aspect of the current state of the anatomical region 202. Here, "updated data" refers to data that is sufficiently current to provide information regarding the current state of the anatomical region 202. Thus, the updated data should preferably be real-time or near real-time data, although for periodic or near-periodic motions having slowly evolving conditions, less current data may be considered updated data.

[0168] The monitoring system 120 may be configured to monitor, for example, the position and / or orientation in space of a boundary region 206 of the anatomical region 202, for example the envelope of the chest of a radiation therapy patient, i.e. the outer contour (body surface) of the chest, for example indicated by skin markers. However, the monitored data may additionally or alternatively include the position and / or orientation in space of an internal structure of the anatomical region, for example the position and / or orientation in space of one or more fiducial markers 208. This data may be received by the monitoring system 120 from an imaging system 122, for example a camera operating in the visible electromagnetic spectrum or an ultrasound imaging system. Alternatively or additionally, 2D images, for example MRI or X-ray projection images, may be obtained by tracking landmarks in the patient, whether fiducial marks or other anatomical landmarks. However, 3D-MRI or 3D-CT systems may not be well suited to be used to provide updated data, as they usually involve relatively long acquisition and reconstruction times, for example high data latency. However, for slowly evolving cyclic or semi-cyclic motions, 3D-MRI and 3D-CT systems may alternatively or additionally be used to provide updated data.

[0169] Preferably, the monitoring system 120 is also arranged to monitor the beam sources 102 and receive updated data regarding the current position and / or alignment of the at least one beam source 102. This data may be received with low latency by the monitoring system 120 from the beam supervision system 130.

[0170] As previously mentioned, the biomechanical model 300 of the anatomical region 202 includes mesh nodes 302 and mesh vertices 304, where each mesh node 302 has its own world-space coordinates:

number

number

number

[0171] Thus, in a graph network system 500 implemented in the control system 110, the trained biomechanical model 300 may be represented as a graph 400, where graph nodes 402 and graph edges 404 are identified to represent mesh nodes 302 and mesh vertices 304, respectively, and the displacement field

number

[0172] Thus, the nodes, edges, and global elements of graph 400 may have the following characteristics:

[0173] [Table 3]

[0174] Dynamic quantities of the system, i.e., state variables

number

[0175] The features received by each graph edge determine the physical interactions and properties from the biomechanical model that are encoded in the network.

[0176] The graph edges are the relative displacement fields

number

number

[0177] The differential operator is a subfunction (f M ) embedded in the graph edges. These functions take edge features (e.g., derivatives and forces) and sender and receiver node features (coordinates, material properties, etc.) as inputs and output higher order derivatives of the nodes. These are then approximated by the dynamic variables

number

[0178] Graph nodes also store functions (f v ) may be embedded.

number

number

number

[0179] The graph processing neural network 504 may process the graph node features, graph edge features, and global features and output the following:

[0180] [Table 4]

[0181] In the updater 506, the dynamic quantity x at time t+1 is then i is obtained by integrating the derivative with respect to time t.

number

number

number

[0182] In the operation of the radiation therapy system 100, the current state M of the trained model 300 t Preferably, the current state M t Establishing the current state M tto data, i.e., reality data, regarding the current position and / or orientation in space of the anatomical region 202. This data may preferably be provided by the monitoring system 120 and may be based, for example, on updated data regarding the boundary region of the anatomical region 202, e.g., the position and / or orientation in space of the envelope of the chest of a radiation therapy patient, i.e., the external contour (body surface) of the chest, and / or the position and / or orientation in space of an internal structure of the anatomical region, e.g., the position and / or orientation in space of one or more fiducial markers.

[0183] If the current state is M t If the current state M of the model 300 is not correlated with the real-world data provided by the monitoring system 120, t may be adjusted or updated to better represent real-world data. If the current state M t If the correlation between the simulated and real-world data is poor, for example below a predetermined limit, the radiation therapy session may be aborted.

[0184] If the current state is M t is found to represent the real data, the estimated future state M t+1 is provided by the graph network system 500, and the position and / or alignment of at least one beam source 102 is determined by the estimated future state M t+1 The illumination target 204 is controlled to track based on the

[0185] The radiation therapy system 100 estimates the future state M t+1 may be configured to be provided intermittently at a period in the range of 20 ms to 100 ms, for example, every 50 ms.

[0186] According to the above embodiment, the current state M t is the current state M of the trained model 300. tEach time the current state M is established, it is correlated to real data. In such a case, the update rate of the monitoring system must be lower than the update rate of the model, for example every 50 ms. According to an alternative embodiment, the current state M t For example, in real data, the frequency is lower, e.g., the current data M t In such an embodiment, the current state M t is mainly the past estimated state of the model, M t-1 , M t-2 , M t-3 The method may be based on at least one of the following:

[0187] The radiation therapy system 100 according to the present disclosure allows prediction of the future position, alignment, and / or deformation of the irradiation target 204, giving the beam supervision system 106 time to react to changes in the state of the irradiation target 204. The radiation therapy system 100 also allows accurate tracking of the irradiation target 204 despite the fact that updated real-world data about the position, alignment, and / or deformation of the irradiation target 204 is not available.

[0188] Since the anatomical region 202 is also modeled to include organs and tissues surrounding the irradiation target 204, movements and deformations of the organs at risk can also be predicted, thus enabling the control system 110 to instruct the beam supervision system 106 to position and align the beam source 102 in such a way that irradiation of such organs can be avoided or at least minimized. Additionally or alternatively, if the estimated future state indicates that there is a risk of the radiation beam intersecting the organ at risk, the control system 110 may be configured to instruct the beam supervision system 106 to block the radiation beam or to stop the corresponding beam source. Blocking the radiation beam may be achieved by applying a gating function to the beam source.

[0189] Additionally or alternatively, the control system 110 may be configured to instruct the beam supervision system 106 to block the radiation beam or stop the corresponding beam source if the estimated future state indicates that the irradiation target is likely to move outside the trajectory of the ionizing radiation beam, for example due to a situation in which the maximum range of the beam source alignment system may not be able to accommodate the predicted movement of the irradiation target.

[0190] As mentioned above, the trained model 300 may advantageously include variables characterizing the planned radiation dose and / or the maximum tolerated radiation dose in or at each mesh node 302 or voxel, and also variables characterizing the accumulated dose in or at each mesh node 302. By receiving updated information regarding the position and alignment of the beam source, and thus information indicating where radiation energy has been deposited in the trained model 300, the control system 110 can update each estimated future state M of the trained model. t+1 , the accumulated radiation dose within or at each mesh node 302 can be updated, thus enabling the control system 110 to instruct the beam supervision system 106 to position and align the beam source 102 such that the irradiation target 204 receives the planned radiation dose without exceeding the maximum allowable radiation dose to other parts of the anatomical region 202.

[0191] FIG. 8 illustrates an embodiment of a method for establishing a biomechanical model of an anatomical region of a radiation treatment patient for use in a radiation treatment system, such as for use in the radiation treatment system 100 described above.

[0192] The method includes establishing 802 a point-based or mesh-based model of the modeled anatomical region. The point-based or mesh-based model (300) may be, for example, the mesh-based model 300 described above.

[0193] The method further comprises a step 804 of binding the static and / or dynamic features to points or mesh nodes of the point-based or mesh-based model. The points or mesh nodes of the point-based or mesh-based model may be, for example, the above-mentioned mesh nodes 302 of the mesh-based model 300. The static and / or dynamic features include world-space coordinates of the points or mesh nodes. For example, the world-space coordinates of the points or mesh nodes may be, for example, the above-mentioned world-space coordinates of the mesh nodes 302.

number

number

[0194] For each point or mesh node, the static and / or dynamic features also include at least one feature characterizing a body tissue within the anatomical region. The at least one feature characterizing a body tissue within the anatomical region may include at least one variable characterizing an interaction between the anatomical region and the ionizing radiation beam, such as at least one of linear stopping power, relative stopping power, radio density, and radiopacity.

[0195] The method further includes representing the model as a graph in a graph neural network system 806. For example, step 806 may include representing the model 300 described above as a graph in the graph neural network system 500.

[0196] The method also includes training a model for periodic or semi-periodic motion of the anatomical region 808. For example, step 808 may include training the model 300 for periodic or semi-periodic motion of the anatomical region 202, as described above.

[0197] Representing the model as a graph in a graph neural network system step 806 may include establishing graph nodes, graph edges, and graph global elements having features that reflect the static and dynamic variables of the biomechanical model, such as the features of model 300 described above.

[0198] For example, if the model 300 described above is represented as a graph in the graph neural network system 500 described above, the graph network system 500 may be configured to include an input 502, a graph processing neural network 504, an updater 506, and an output 508, as described above, and the graph processing neural network 504 may include an encoder 510, a processor 512, and a decoder 514. The encoder 510 derives the current state M of the mesh-based biomechanical model 300 from the input 502. t , and the current state M of the biomechanical model 300 is t from the input graph G 0 The processor 512 may be configured to encode the input graph G 0 , and output graph G with updated node and edge embeddings. M The decoder 514 may be configured to provide an output graph G that extracts output features corresponding to the static and dynamic variables of the biomechanical model 300. M , and the updater 506 updates the output features to estimate the future state M t+1 Generate and estimate the state M t+1 at output 508.

[0199] The step 808 of training a model for periodic or semi-periodic motion of an anatomical region includes the step of: t+1 and a real data set of the anatomical region corresponding to the modeled anatomical region. The real data set may include at least one of a four-dimensional computed tomography (4D-CT) scan, a four-dimensional magnetic resonance imaging (4D-MRI), and ultrasound imaging. The real data set may include a progression of images covering the modeled anatomical region during a respiratory cycle.

[0200] The point-based or mesh-based model may incorporate at least one of a linear elastic model and a hyperelastic model adapted to the physical properties of the anatomical region.

[0201] The step 802 of establishing a point-based or mesh-based model of the modeled anatomical region may comprise a first sub-step 802a of dividing the anatomical region into at least a first anatomical sub-region displaying a first set of physical properties and a second anatomical region displaying a second set of physical properties different from the first set of physical properties, a second sub-step 802b of assigning, for the first anatomical sub-region, to the point-based or mesh-based model a first point-based or mesh-based sub-model adapted to the first set of physical properties, and a third sub-step 802c of assigning, for the second anatomical sub-region, to the point-based or mesh-based model a second point-based or mesh-based sub-model adapted to the second set of physical properties. If the first and second anatomical sub-regions share a common boundary, the sub-steps 802b and 802c may comprise assigning common boundary conditions at the common boundary to the first and second point-based or mesh-based sub-models.

[0202] An embodiment of a computer-implemented method for planning at least a part of a radiation therapy treatment to be performed on an anatomical region of a radiation therapy patient is shown in Fig. 9. The method includes a first step 902 of establishing a biomechanical model of the anatomical region according to the method described above with reference to Fig. 8, and a second step 904 of assigning to each point or mesh node at least one of a planned dose and a maximum tolerated dose, the model enabling the absorbed dose at different points or mesh nodes to be compared with the planned dose or the maximum tolerated dose during a subsequent radiation therapy session. As described above, a radiation therapy system in which the model is implemented can then control dose delivery and beam alignment based on the difference between the planned dose and the absorbed dose such that the irradiation target receives the planned radiation dose without exceeding the maximum tolerated radiation dose for other parts of the anatomical region.

[0203] In the preceding description, various aspects of the apparatus according to the present invention have been described with reference to exemplary embodiments. For purposes of explanation, specific numbers, systems and configurations have been set forth to provide a thorough understanding of the apparatus and its workings. However, this description is not intended to be construed in a limiting sense. Various modifications and variations of the exemplary embodiments, as well as other embodiments of the apparatus, will be apparent to those skilled in the art to which the disclosed subject matter pertains, and are deemed to be within the scope of the present invention as defined by the following claims.

Claims

1. 1. A method for establishing a biomechanical model (300) of an anatomical region (202) of a radiation therapy patient (200) for use in a radiation therapy system (100), the method comprising: (a) establishing a point-based or mesh-based model (300) of said anatomical region (202); (b) associating static and / or dynamic features with points or mesh nodes (302) of said point-based or mesh-based model (300), said features being associated with world-space coordinates of said points or mesh nodes (302); [0063] and state variables representing the spatial-temporal displacement of said anatomical region (202). [0,64] and, for each point or mesh node (302), at least one feature characterizing a body tissue within said anatomical region; (c) representing the model (300) as a graph in a graph neural network system (500); (d) training the model (300) for periodic or semi-periodic motion of the anatomical region (202); The method includes:

2. The method of claim 1 , wherein the at least one feature characterizing body tissue within the anatomical region includes at least one variable characterizing an interaction between the anatomical region (202) and an ionizing radiation beam.

3. 3. The method of claim 2, wherein the at least one feature characterizing an interaction between the anatomical region (202) and the ionizing radiation beam includes at least one of linear stopping power, relative stopping power, radio density, and radiopacity.

4. Representing the model (300) as a graph in a graph neural network system (500) includes establishing graph nodes, graph edges, and graph global elements having characteristics reflecting static and dynamic variables of the biomechanical model (300); The graph network system (500) comprises an input (502), a graph processing neural network (504), an updater (506), and an output (508), the graph processing neural network (504) comprising an encoder (510), a processor (512), and a decoder (514); The encoder (510) derives the current state (M t ) of the biomechanical model (300) and t ) to the input graph (G 0 ) The processor (512) calculates the input graph (G 0 ) and output graph (G M ) configured to provide The decoder (514) extracts output features corresponding to the static and dynamic variables of the biomechanical model (300) from the output graph (G M ) The updater (506) updates the output features to estimate the future state (M t+1 ), and the estimated state (M t+1 4. The method of claim 1, further comprising providing a signal to said output (508).

5. Training the model (300) for periodic or semi-periodic motion of the anatomical region (202) may be performed using the estimated states (M t+1 5. The method of claim 4, comprising applying a loss function to a model of the anatomical region (202) and to a real data set of the anatomical region corresponding to the modeled anatomical region (202).

6. The method of claim 5, wherein the real-world data set comprises at least one of a four-dimensional computed tomography (4D-CT) scan, a four-dimensional magnetic resonance imaging (4D-MRI), and ultrasound imaging.

7. The method of claim 5 or 6, wherein the real data set comprises a spread of images covering the modeled anatomical region (202) during a respiratory cycle.

8. The method of any one of claims 1 to 7, wherein the point-based or mesh-based model incorporates at least one of a linear elastic model and a hyperelastic model adopted to the physical properties of the anatomical region (202).

9. dividing the anatomical region (202) into a first anatomical sub-region (202a) displaying a first set of physical properties and a second anatomical sub-region (202b) displaying a second set of physical properties different from the first set of physical properties; assigning, for the first anatomical sub-region (202a), to the point-based or mesh-based model (300), a first point-based or mesh-based sub-model adapted to the first set of physical properties; and assigning, for the second anatomical sub-region (202b), to the point- or mesh-based model (300), a second point- or mesh-based sub-model adapted to the second set of physical properties.

10. 10. The method of claim 9, wherein the first and second anatomical sub-regions share a common boundary, the method comprising assigning common boundary conditions at the common boundary to the first and second point-based or mesh-based sub-models.

11. 1. A computer-implemented method for planning at least a portion of a radiation therapy treatment to be performed on an anatomical region of a radiation therapy patient, comprising: Establishing a biomechanical model (300) of said anatomical region (202) according to any one of claims 1 to 10; and assigning to each point or mesh node at least one of a planned dose and a maximum allowed dose.

12. at least one beam source (102) arranged to provide a beam of ionizing radiation (104) to an irradiation target (204) located within an anatomical region (202); a beam supervision system (106) configured to control the position and / or alignment of the at least one beam source (102); a monitoring system arranged to monitor said anatomical region (120); A radiation therapy system (100) comprising: a control system (110) comprising a computer-implemented graph network system (500) on which a model (300) of the anatomical region (202) trained for periodic or semi-periodic motion of the anatomical region (202) is implemented; The graph network system (500) calculates the current state (M t ), and further configured to establish the current state (M t ) based on the estimated future state (M t+1 ) configured to establish The control system (110) calculates the estimated future state (M t+1 ) a beam supervision system (106) configured to instruct the beam supervision system (106) to control the ionizing radiation beam (104) of the at least one beam source (102) based on the beam supervision system (106) and the at least one beam source (102).

13. The control system (110) calculates the estimated future state (M t+1 13. The radiation therapy system (100) of claim 12, configured to instruct the beam supervision system (106) to control the ionizing radiation beam (104) of the at least one beam source (102) based on a target position (204) and cause the at least one beam source (102) to track the irradiation target (204).

14. The control system (110) calculates the estimated future state (M t+1 ) based on the estimated future state (M t+1 14. The radiation therapy system (100) of claim 12, further configured to instruct the beam supervision system (106) to block the ionizing radiation beam (104) of the at least one beam source (102) or to stop the at least one beam source (102) if the at least one beam source (102) is configured to instruct the beam supervision system (106) to block the ionizing radiation beam (104) of the at least one beam source (102) or to stop the at least one beam source (102) when the at least one beam source (102) is configured to instruct the beam supervision system (106) to block the ionizing radiation beam (104) of the at least one beam source (102) if the at least one beam source (102) is configured to instruct the beam supervision system (106) to stop the at least one beam source (102) when the at least one beam source (102) is configured to instruct the beam supervision system (1

15. The control system (110) calculates the estimated future state (M t+1 ) based on the estimated future state (M t+1 15. The radiation therapy system (100) of claim 12, wherein said beam supervision system (106) is configured to instruct said beam supervision system (106) to block said ionizing radiation beam (104) of said at least one beam source (102) or to stop said at least one beam source (102) when said at least one beam supervision system (106) detects a risk that a trajectory of said ionizing radiation beam (104) may intersect with an organ at risk.

16. The trained model (300) is a point-based or mesh-based model (300) comprising static and / or dynamic features associated with points or mesh nodes (302) of the point-based or mesh-based model (300), the features being associated with world-space coordinates of the points or mesh nodes. [0.65] and state variables representing the spatial-temporal displacement of said anatomical region (202). [0.66] The radiation therapy system (100) of any one of claims 12 to 15, comprising:

17. The radiation therapy system (100) of any one of claims 12 to 16, wherein the graph network system (500) is configured to represent the trained model (300) as a graph (400) including graph nodes (402), graph edges (404), and graph global elements having features reflecting static and dynamic variables of the trained model (300).

18. The graph network system (500) calculates the established current state (M t 18. The radiation therapy system (100) of any one of claims 12 to 17, configured to associate a signal from the radiation therapy system (100) with the data received from the monitoring system (120).

19. 20. The radiation therapy system of claim 18, wherein the monitoring system (120) is configured to monitor at least one of a spatial position and / or orientation of a boundary region (206) of the anatomical region (202) and a spatial position and / or orientation of a fiducial marker (208) within the anatomical region (202).

20. the variables include at least one variable characterizing an interaction between the anatomical region (202) and a beam of ionizing radiation, and at least one variable characterizing at least one of a planned radiation dose, a maximum tolerable radiation dose, and an accumulated radiation dose; Each estimated future state (M t+1 ) the graph network system (500) is configured to update the variables characterizing the accumulated radiation dose; 20. The radiation therapy system of claim 12, wherein the control system (110) is configured to instruct the beam supervision system to control a position and / or alignment of the at least one beam source such that the irradiation target (204) receives a planned radiation dose without exceeding the maximum allowable radiation dose for other parts of the anatomical region (202).

21. The radiation therapy system (100) of any one of claims 12 to 20, wherein the monitoring system (120) is configured to monitor the beam supervision system (106).

22. The radiation therapy system (100) of any one of claims 12 to 21, wherein the trained model (300) of the anatomical region (202) is established according to any one of claims 1 to 10.

Images